Effect of Sternal Sensor Location on Seismocardiography for Severe Aortic Stenosis Detection

Aortic stenosis (AS) is a common valvular disease in older adults, and seismocardiography (SCG), which records chest-surface vibrations associated with cardiac mechanical activity, has been investigated as a noninvasive method for cardiac assessment. This study evaluated how sternal sensor location affects SCG analysis in a dataset of 160 subjects, including 101 patients with severe AS and 59 age- and sex-comparable controls. Each subject was measured at the upper and lower sternum with a custom device that recorded single-lead electrocardiography and three-axis chest acceleration. Electrocardiography was used for cardiac-cycle gating, and time-frequency features were extracted from cardiac cycles selected with a dynamic time warping quality-control procedure. A gradient boosting classifier was used in position-label tests that distinguished upper from lower sternum measurements and in disease-label tests that distinguished severe AS from controls. Model performance was evaluated using leave-one-subject-out cross-validation with position-level and patient-level aggregation. For the position-label task, upper and lower sternum measurements were distinguishable in controls, AS subjects, and the combined cohort, with area under the receiver operating characteristic curve values of 0.919, 0.980, and 0.966, respectively. For the disease-label task, performance was highest when training and testing were both performed at the upper sternum, yielding sensitivity 0.960, specificity 0.949, accuracy 0.956, and area under the receiver operating characteristic curve 0.983. Cross-position transfer reduced performance substantially. These results show that sensor location affects SCG analysis and should be considered when developing and evaluating SCG-based methods for AS detection.